A new hybrid financial time series prediction model. (October 2020)

• Record Type: Journal Article
• Authors: Alhnaity, Bashar; Abbod, Maysam
• Is Part Of: Engineering applications of artificial intelligence. Volume 95(2020)
• Publication Date: 2020-10
• Subjects: Neural networks -- Support vector machine -- Genetic algorithm -- Ensemble empirical mode decomposition -- Financial time series; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2020.103873
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Due to the characteristics of financial time series, such as being non-linear, non-stationary and noisy, with uncertain and hidden relationships, it is difficult to capture its non-stationary state and to accurately describe its moving tendency. This is also a consequence of using a single approach to artificial intelligence, and other techniques that have been conventionally used. Therefore, those participating in financial markets, along with researchers, have paid a great deal of attention to tackling this problem. Hence, several approaches have been developed to alleviate the influence of inherent characteristics. However, the noise characteristic can refer to the unavailability of information, which affects how financial markets behave, as well as captured prices in both the past and the future. Therefore, the prediction of stock prices and detecting their noise is considered a very challenging financial topic. This paper adopts a novel three-step hybrid intelligent prediction model that combines a collection of intelligent modelling techniques and a feature extraction algorithm. At first, ensemble empirical mode decomposition is applied to the original data, as to facilitate model fitting to them. Then neural network and support vector regression is proposed individually for modelling the extracted features. Finally, a weighted ensemble average using a genetic algorithm to optimise and determine the weight is proposed for establishing a unified prediction. Experimental results are presented which illustrate the excellent performance of the proposed approach, and that is significantly outperforming the existing models, in terms of error criteria such as MSE, RMSE and MAE.